Postdoctoral Fellowship: OCE-PRF: Using machine learning to investigate temporal dynamics of methane seep fauna at the Ocean Observatories Initiative (OOI) Regional Cabled Array

甲烷水合物渗漏是一种独特的化学合成海洋环境，生物利用海底的化学物质而不是阳光来创造能量。它们分布在世界各地的大陆边缘（仅俄勒冈州和华盛顿沿海就有1 300多个），提供重要的服务，如以甲烷水合物形式储存能源，为具有重要商业价值的鱼类种群提供栖息地，并为区域生物多样性和生产力作出贡献。然而，目前尚不清楚这些关系是否是由于化学合成生产的生产力提高，海底栖息地复杂性增加或两者兼而有之。也不知道这些栖息地使用者在什么样的时间尺度下运作：它们是个体的随机一次性互动，还是更广泛人群的定期、可预测的互动？因此，很难量化和充分描述甲烷渗漏提供的服务，这是成功管理这些潜在的关键栖息地，能源和碳汇的重要信息。这项研究将通过使用海洋观测站倡议（OOI）区域电缆阵列在南海瑟薇岭（俄勒冈州海岸附近的甲烷水合物渗漏）收集的十多年独特的图像和视频时间序列来探索这些问题。 该项目还将涉及机器学习管道的开发，这将使研究人员和公众更容易获得OOI图像的未来工作，因为目前手动注释图像的方法非常耗时。此外，该项目将支持一名早期职业博士后研究员，为两名本科生提供参与研究的机会，并制作一个面向广大本科生观众的关于海洋技术和基于图像的海洋研究的课堂模块。该项目将开发多个分类单位探测器作为机器辅助注释管道的一部分，用于处理OOI区域有线阵列收集的估计37 TB的图像和视频。为该项目开发的管道将有助于快速发展的海洋自动成像领域，其目的是使每天在世界海洋中收集的大量图像以及历史上收集的数据更容易用于科学研究。可见的动物，细菌垫和地质结构将在图像中注释，量化随着时间的推移该地点社区的丰度和多样性的变化。然后，这些数据将用于调查甲烷渗漏处海底巨型动物群落的时间轨迹，并研究环境条件如何影响该群落。洞察力将帮助我们理解这些环境中的每日、每月和每年的周期，这将更好地理解基线时间动态。总的来说，该项目将有助于甲烷水合物渗漏生态动力学的知识，为这些生物多样性，高碳，该项目由地球科学理事会共同资助，以支持人工智能/ML在地球科学中的进步。该奖项反映了NSF的法定使命，并通过使用基金会的智力价值进行评估而被认为值得支持和更广泛的影响审查标准。

Methane hydrate seeps are unique chemosynthetic marine environments where organisms use chemicals from the seafloor rather than sunlight to create energy. They are found on continental margins worldwide (with more than 1,300 off the coast of Oregon and Washington alone) providing important services such as energy storage in the form of methane hydrates and habitat for commercially important fish stocks, as well as contributing to regional biodiversity and productivity. However, it is still unclear whether these relationships are due to enhanced productivity from chemosynthetic production, increased seafloor habitat complexity, or a combination thereof. Nor is it known what timescales these habitat users operate under: are they random one-off interactions by individuals, or regular, predictable interactions by a broader population? Therefore, it is difficult to quantify and fully describe the services methane seeps provide, which is vital information for successful management of these potentially critical habitats, energy resources, and carbon sinks. This research will explore these questions by using a unique time series of imagery and video collected by the Ocean Observatories Initiative (OOI) Regional Cabled Array over more than a decade at Southern Hydrate Ridge, a methane hydrate seep off the coast of Oregon. The project will also involve the development of machine learning pipelines which will make future work with OOI imagery more accessible for researchers and the public, since current, manual methods of annotating imagery are incredibly time consuming. Additionally, this project will support an early career post-doctoral researcher, give two undergraduate students an opportunity to participate in research, and produce a classroom module - geared to a broad undergraduate student audience - on marine technology and imagery based marine research.Specifically, this project will develop multi-taxon detectors as a part of a machine assisted annotation pipeline to process an estimated 37 TB of imagery and video collected by the OOI Regional Cabled Array. The pipeline developed for this project will contribute to a fast-moving field of marine automated imaging which aims to make the large amounts of imagery collected in the world’s oceans every day, as well as historically collected data, more readily usable for scientific research. Visible animals, bacterial mats, and geologic structures will be annotated in the imagery, quantifying the changes in abundances and diversity of the community at the site over time. This data will then be used to investigate the temporal trajectory of the benthic megafauna community at the methane seep and examine how the community is influenced by environmental conditions. Insights will help us understand daily, monthly, and yearly cycles in these environments, which will create better understanding of baseline temporal dynamics. Overall, this project will contribute knowledge of methane hydrate seep ecologic dynamics, providing vital information for characterization and management of these biodiverse, high-carbon, and energy-rich habitats.This project is co-funded by the Directorate for Geosciences to support AI/ML advancement in the geosciences.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.